Should you trust AI to run your investment portfolio?

How AI is being used in the investment world?

The fund management world has already been employing computer-driven models for many years in the form of ‘quantitative’ or ‘systematic’ funds. AI is likely to become a more important part of the research processes used by fund managers in future, used to rapidly process vast amounts of data to provide insights to aid decision-making. However, regulation is likely to stand in the way of fully autonomous AI-managed funds emerging in the near term – and, for the time being, this may be a good thing.

A recent experiment testing the performance of a list of 38 shares identified by OpenAI’s ChatGPT versus 10 popular funds over 10 weeks has garnered plenty of media attention, as the AI-generated list has reportedly delivered higher returns. It should be pointed out though, that while the ChatGPT ‘portfolio’ was entirely comprised of large companies listed on the US market, nine of the 10 funds it was compared against were index tracker funds and only one of which was a US equity fund. The list it was compared against included two funds that included exposure to bonds and two trackers that only invest in the UK market, so the comparison was not like-for-like. Not to mention, real funds have costs which a fantasy portfolio does not.

Bestinvest's analysis on robot stock selectors

To delve deeper into the topic of whether people should use AI chatbots to select their investments, a national newspaper recently sought our help in assessing the stock lists generated by two leading AI chatbot platforms, ChatGPT (developed by OpenAI) and Bard (developed by Google), asking each to come up with a share portfolio.

It is important to recognise that a real fund is much more than a list of shares, as asset allocation – how the portfolio is spread across different holdings and sectors - is also a crucial factor in long-term returns. The experiment does not consider the running costs (even an AI-generated portfolio needs to be bought, with all the associated brokerage costs). Equally, ‘real’ fund managers will not simply buy and hold, but will shape a portfolio around different market conditions and the fortunes of individual companies.

Nevertheless, it is interesting to note the type of companies picked by the robot stock selectors. While there was a 42% overlap between the two lists, including heavyweights such as Alphabet, Amazon, Berkshire Hathaway, Microsoft and Visa, there were also notable differences. Apple, Disney and McDonald’s do not appear on the ChatGPT list, while Bard has no place for Nestle, Netflix or Taiwan Semiconductor.

Geographic concentration

Notably, Bestinvest’s analysis found that both AI chatbots only served up lists of very large companies with shares listed on the US equity markets. All of Bard’s picks were S&P 500 Index companies, while ChatGPT’s found a place for large European and Asian companies with dual listings in the US. The US is undoubtedly a market which includes many fantastic businesses, but it does not have a monopoly on great companies, and it would be risky to focus on just one market.

This reveals some biases in the algorithms: a lot of data is publicly available on large US companies, their shares are traded regularly, and they are widely covered by research analysts. Inevitably, the AI stock picker assigns them more importance, but it means many more opportunities are neglected. The best returns can be made by uncovering hidden gems with the potential to become the Amazon or Microsoft of the future – investors won’t find those here.

Sector exposure

While both lists of stocks were narrowly focused on very large companies with US listings, there were major differences in their exposure to different industry sectors both compared to the US market (S&P 500) and each other.

For example, companies in the materials, utilities and real estate sectors didn’t appear on either list, while ChatGPT had no exposure to energy. This would have exerted a drag on returns over the past 12 months when energy companies have soared. Instead, the ChatGPT list has very high exposure to technology shares - 13 out of 38 (the Bard list only has eight tech stocks). Thinking machines can therefore come to very different conclusions, much as human fund managers will form different views depending on their approaches to investing. 

The Bard list has a much higher exposure to financials than ChatGPT, with seven stocks (21%) compared to four in the ChatGPT list (10.5%). Only two banks made either list – with bulge bracket Wall Street banks Goldman Sachs and Wells Fargo appearing in the Bard list. ChatGPT had high exposure to Healthcare (18.4%) and industrials (10.5%) compared to Bard.

Performance

Performance measured in an experiment lasting just a few weeks is statistically insignificant and should not be relied upon. We were therefore asked to look at how both lists would have fared if you had purchased these shares with equal weightings in each stock 12 months ago – a tough period for US equities. We estimate that in Pound terms, the Bard list would have delivered a total return of 1.47% in the 12 months to the end of April. This underperformed the 2.55% return from the S&P 500 Index over the same period and the 3.61% return from the MSCI World Index.

The ChatGPT list would have returned a higher 3.22% over the same period, better than the S&P 500 but still behind the MSCI World Index. However, this was down to the contribution of Netflix, which rallied 73% from the end of April 2022 after seeing its share price crash from their all-time high in October 2021. Without that lift, the ChatGPT portfolio would have returned a less convincing 1.29%.

Neither list would have outpaced the UK market’s largest 100 companies, which returned 8.2% over the year. As a reminder, both AI chatbots only chose companies with shares listed on the US markets.

Interestingly, had someone invested just in the 16 shares that appeared on both AI-generated lists, we estimate that the 12-month return would have been a meagre 0.53% - the worst outcome of all. It should be pointed out that none of these return figures factor in any costs, such as account or dealing fees, which in the real world an investor would incur were they to turn an AI set of investment ideas into an actual portfolio.

Should you trust AI with your investments?

Our conclusion is that while AI is going to have a major, disruptive impact on many walks of life – including the world of investing, where the ability to process vast amounts of data has huge potential to provide research ideas and insights – people should not blindly adopt the output of AI chatbots.

When creating a portfolio for the long term, it's crucial to think about risk and make sure you diversify across various asset classes, markets, and sectors. You can achieve this by selecting low-cost index funds, carefully chosen actively managed funds led by experienced managers with good track records, a combination of both, or building a portfolio of direct share investments.

Performance data was derived from Morningstar and represents total return to end April 2023.

Important information

By necessity, this briefing can only provide a short overview and it is essential to seek professional advice before applying the contents of this article. This briefing does not constitute advice nor a recommendation relating to the acquisition or disposal of investments. No responsibility can be taken for any loss arising from action taken or refrained from on the basis of this publication. Details correct at time of writing.

The value of an investment may go down as well as up and you may get back less than you originally invested. Past performance is not a guide to future performance.

The opinions expressed are made in good faith but are subject to change without notice.